1 tl;dr

1.1 recommendations

  • performance of these tags was probably sufficient that further deployments could be justified.
  • alternative programming configurations should be considered.
  • sample size and analysis types should be considered before deploying additional fastlocs.

1.2 summary

  • It is possible to collect at least 14 days of near continuous 5 minute time-series depth data simultaneously with fastloc positions.
  • Average fastloc fixes tx to satellite per day were low in this configuration, however, ranging from <1 to 4 for the tags with time-series data enabled.The fastloc only tag faired better with an average of 7.6 fastloc fixes per day tx to satellite.
  • Fastloc positions did increase the accuracy and precision in of modeled tracks in almost all cases, though the average magnitude of improvement in accuracy over an argos only tag was modest ranging from about ~2-5 kilometers.
  • Of the 4 tags collecting dive data, 1 tag was a near total losses for dive data and fastloc data with a very short deployment and low tx rate. 1 Tag had a high incidence of gaps and 2 tags had nearly complete dive records.
  • Both tags with nearly complete dive records suffered pressure transducer failures after the 14 days of dive data collection.

2 BACKGROUND

We are assessing the utility and feasibility of instrumenting Ziphius cavirostris off Cape Hatteras with fastloc capabale wildlife computer splash tags for the atlbrs.

We have deployed a total of 5 fastloc capable tags between 2018 and 2023. The first tag deployed in 2018 was programmed to only collect fastloc data and no dive data. The follow 4 tags were programmed to collect both time-series dive data and fastloc positions. Settings details can be found below.

Each fastloc data message consists of the data required to calculate a single fastloc position. As a comparison at 5-minute sampling period a single time-series data message spans 4 hours of recording. Therefore if we attempt to generate even hourly fastloc positions, our data queue for fastloc quickly surpasses the bandwidth requirements for dive data alone. Therefore it is unlikely even if we forego all other data streams that we would be able to densely sample the movement of an instrumented animals during the time scale of a single treatment event which last from 30 minutes to 1 hour and with a window of highest interest extending 24 hours past the conclusion of the treatment.

Therefore, there are two main questions regarding the use of fastloc tags in the atlbrs:

  1. Can the programming be tuned to produce a satisfactorily continuous dive data record along side some fastloc positions.
  2. Is the number of fastloc positions produced (and received) by this tuning sufficient to increase precision and accuracy of modeled positional tracks above what ARGOS alone already provides at a relevant time scale to our treatments. Obviously, decreased error in modeled tracks could impact RL modeling as well as avoidance beahvior analyses.

3 programming

We used two tag programming configurations: (1) a fastloc only configuration, and (2) a time-series and fastloc configuration. Programming highlights:

fastloc only configuration fastloc and time-series configuration
fastloc:
- 2/hr max successful fastloc attempts
- 4/hr max failed fastloc attempts
- 96/day overall max fastloc attempts
- 0-23 fastloc attempt hours
- unlimited collection period
fastloc:
- 1/hr max successful fastloc attempts
- 4/hr max failed fastloc attempts
- 40/day overall max fastloc attempts
- 3-4, 9-10, 15-16, 21-22 fastloc attempt hours
- 21 day collection period
argos tx:
- 470 daily tx
- 24 hours initial tx, then:
- 0-3, 7-23 tx hours
argos tx:
- 470 daily tx
- 24 hours initial tx, then:
- 0-3, 8-11, 12-23 tx hours
time-series:
- depth
- 5-minute sampling peirod
- 14 day collection period
buffer and tx priority:
- 2 day data buffer
- high priority fastloc
buffer and tx priority:
- 100 day data buffer
- high priority time-series and fastloc

4 instrumentation details

ZcTag077_DUML

  • catalog id:Zca_015_HAT
  • previously tagged ZcTag039
  • adult male
  • first seen 2015
  • fastloc only programming

ZcTag143_DUML

  • catalog id: Zca_098_HAT
  • previously tagged: ZcTag102_DUML
  • adult male
  • first seen 2020
  • perhaps eroded or missing teeth
  • fastloc and time-series programming

ZcTag144_DUML

  • catalog id: ???
  • adult male
  • fastloc and time-series programming

ZcTag145_DUML

  • catalog id: ???
  • unknown age unknown sex
  • fastloc and time-series programming

ZcTag146_DUML

  • catalog id: Zca_072_HAT
  • unknown age male (genetically determined)
  • first seen 2019
  • fastloc and time-series programming

5 locations

ZcTag077_DUML

quality 0 1 2 3 A B gps ndays
count 55 9 3 1 28 92 179 23.7
per day 2.3 0.4 0.1 0 1.2 3.9 7.6

ZcTag143_DUML

quality 0 1 2 3 A B gps Z ndays ndays gps
count 231 68 19 1 118 323 81 2 67.3 21
per day 3.4 1 0.3 0 1.8 4.8 3.9 0

ZcTag144_DUML

quality 0 1 2 3 A B gps ndays ndays gps
count 83 21 6 1 71 217 47 37.9 21
per day 2.2 0.6 0.2 0 1.9 5.7 2.2

ZcTag145_DUML

quality 0 1 2 A B gps ndays ndays gps
count 20 2 2 8 35 3 5.2 5.2
per day 3.8 0.4 0.4 1.5 6.7 0.6 1

ZcTag146_DUML

quality 0 1 2 3 A B gps ndays ndays gps
count 363 113 33 10 117 307 65 68.8 21
per day 5.3 1.6 0.5 0.1 1.7 4.5 3.1

6 failed versus successful fastlocs

7 dive data and health status

PROVISIONAL plot status cutoffs and bat voltage. Cutoffs are in dotted red lines. These are either:

  1. healthy tags: the last status messages as a conservative cutoff and on tags that are still actively transmitting this is likely to increase
  2. unhealthy pressure transducer tags: if there are 2 status messages (crc’d) where the | zerodepth | > 10 then this is the last good status message before the 1st of those two bad messages.

We have settled on a more complicated cutoff system to capture a little more very likely good data. For detail please see DATAPREP_sattag_processing. But this gives you a general idea of approx. cutoffs and pressure transducer health.

8 impact on accuracy and precision

To estimate accuracy and precision, I fit movement models to the positional data using aniMotum, the successor to foieGras by Ian Jonsen. A baseline model was fit using only the ARGOS positions and a set of models fit with the ARGOS positions and all fastloc positions save for one iterating over each fastloc position. The baseline model was used to predict the position at the time point for each fastloc position, and the leave-one-out models were each used to predict the position at the time point of the particular fastloc position that was dropped from the fitting. Treating the fastloc positions as truth, distance between these positions and the predicted positions from the models were calculated to assess accuracy and precision. Due to the vast difference in average error between even relatively poor fastloc fixes and ARGOS fixes, I felt treating fastloc as truth was an acceptable approximation, which appears also in the literature. Note though that some fastloc positions can have larger errors and so more filtering might be necessary in some cases.

The models were fit using correlated random walks. Data were prepared by removing points far outside the study area, described by the box {x1, x2, y1, y2} = {-82, -62, 30, 42}. Additional a speed filter was applied using a maximum speed of 10 kilometers per hour.

All distances below are measured in kilometers.

deployid mean0 mean_gps sd0 sd_gps
ZcTag077_DUML 7.98 2.68 5.29 2.75
ZcTag143_DUML 10.69 6.39 10.99 6.96
ZcTag144_DUML 7.07 5.37 4.18 3.34
ZcTag146_DUML 7.68 5.64 4.34 4.43